topographic mapping of vmh → arcuate nucleus microcircuits and their reorganization by fasting

Topographic mapping of VMH-arcuate nucleusmicrocircuits and their reorganization by fasting

Scott M Sternson1, Gordon M G Shepherd2 & Jeffrey M Friedman1

In the hypothalamic arcuate nucleus (ARC), pro-opiomelanocortin (POMC) neurons inhibit feeding and neuropeptide-Y (NPY)

neurons stimulate feeding. We tested whether neurons in the ventromedial hypothalamic nucleus (VMH), a known satiety center,

activate anorexigenic neuronal pathways in the ARC by projecting either excitatory synaptic inputs to POMC neurons and/or

inhibitory inputs to NPY neurons. Using laser scanning photostimulation in brain slices from transgenic mice, we found that

POMC and NPY neurons, which are interspersed in the ARC, are nevertheless regulated by anatomically distinct synaptic inputs.

POMC neurons received strong excitatory input from the medial VMH (mVMH), whereas NPY neurons did not and, instead,

received weak inhibitory input only from within the ARC. The strength of the excitatory input from the mVMH to POMC neurons

was diminished by fasting. These data identify a new molecularly defined circuit that is dynamically regulated by nutritional state

in a manner consistent with the known role of the VMH as a satiety center.

The hypothalamus regulates food intake and body weight1. Classicbrain lesioning experiments led to a ‘two-center’ hypothesis inwhich the ventromedial hypothalamic nucleus (VMH) signalssatiety2 and the lateral hypothalamus signals hunger3 (Fig. 1a).Although the lesions in these studies also affected surroundingbrain regions, precise genetic disruption of the VMH by a mutationin the gene encoding steroidogenic factor-1 has confirmed theimportance of the VMH in maintaining energy homeostasis4. Recently,numerous molecular and genetic experiments have also identified thehypothalamic arcuate nucleus (ARC) as a critical regulator ofenergy balance5. Within the ARC, discrete populations of NPYand POMC neurons sense hormonal and metabolic signals andactivate homeostatic responses that favor energy storage or energydissipation6–9. In addition to sensing peripheral metabolic signals,these key, molecularly defined regulatory neurons receive synapticinputs from other brain regions10,11. However, the sources ofmany of these inputs have not been determined. Based on data fromlesioning studies, we reasoned that some of these inputs might comefrom the VMH.

Although the VMH was first identified as a satiety center more than60 years ago, the mechanism by which it regulates satiety and bodyweight is largely unknown2. In these studies, we considered thepossibility that the VMH regulates energy balance by modulating theactivity of discrete populations of neurons in the ARC. Specifically, weset out to delineate the functional connections between the VMH andthe ARC with the aim of integrating the classic neuroanatomic studiesof Hetherington and Ranson2 with the more recent identification ofmolecularly defined neuronal populations in the ARC. The VMH isadjacent to the ARC (Fig. 1a), but synaptic inputs from the VMH to

POMC or NPY neurons in the ARC have not been reported. Althoughinjection of an anterograde neuronal tracer into the VMH does notlabel the ARC, the cell-poor region separating the VMH and the dorsalsurface of the ARC does reveal heavy innervation12 (see Figs. 3 L-O and4G-I in ref. 12). Other studies show that Golgi-stained ARC neuronsproject dendrites as a ‘shell’ along this dorsal ARC surface13, suggestingthat there are potential sites of interaction between the VMH andARC. We thus considered the hypothesis that VMH neuronsactivate anorexigenic neuronal pathways in the ARC by projectingexcitatory synaptic inputs to POMC neurons and/or inhibitory inputsto NPY neurons.

To identify the microcircuits within hypothalamic feeding centers,we used laser scanning photostimulation14–21 (LSPS) to map thedistributions of presynaptic partners in the VMH onto individualneurons in the ARC (Fig. 1b). LSPS measures the origin and strength offunctional synaptic projections to individual neurons within a brainslice by stimulating small clusters of presynaptic neurons with gluta-mate uncaging while simultaneously recording evoked postsynapticcurrents (PSCs) in whole-cell voltage clamp. The reversal potential ofthe evoked current is used to characterize the connection as excitatoryor inhibitory. Uncaging glutamate can give rise to three types ofresponses in the recorded neuron (Fig. 1c). First, a direct responseoccurs when glutamate uncaging occurs near the soma or dendrites ofthe voltage-clamped neuron, directly activating glutamate receptors.Second, a synaptic response (that is, a PSC) occurs when glutamateuncaging activates a presynaptic neuron to fire an action potential,which induces neurotransmitter release, resulting in an evokedsynaptic current in the voltage-clamped neuron (Fig. 1d). Third, noresponse is observed at sites of glutamate uncaging that do not

Received 21 July; accepted 25 August; published online 18 September 2005; doi:10.1038/nn1550

1Howard Hughes Medical Institute and Laboratory of Molecular Genetics, The Rockefeller University, 1230 York Ave., New York, New York 10021, USA. 2Howard HughesMedical Institute and Cold Spring Harbor Laboratory, Cold Spring Harbor, New York 11724. Correspondence should be addressed to J.M.F. ([email protected]).

1356 VOLUME 8 [ NUMBER 10 [ OCTOBER 2005 NATURE NEUROSCIENCE

ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

contain neurons connected to the voltage-clamped neuron. Note thatunder the conditions used here, only monosynaptic inputs wereobserved (see below).

Overall, LSPS provides an unbiased means for locating presynapticneurons and has been well validated as a method for discovery of newcircuit connections15,17. For this purpose, LSPS has advantages relativeto other approaches such as electrical stimulation, which is limited byuncertainty about the origin of connections because axons of passageare also stimulated, and paired recordings, which are inefficient due tolow connection probabilities and the corresponding paucity of con-nections that can be sampled per experiment. Previously, LSPS hasbeen used to map cortical15–17,20 and hippocampal18 circuits, but it hasnot yet been applied to mapping hypothalamic circuits.

RESULTS

In this study, glutamate uncaging was repeated at 256 discrete pointsacross a brain slice (300 mm thick) that included the ARC and VMH,

allowing a 1.3 mm2 region of the neuropil to be rapidly assayed for sitesof connectivity (Fig. 1b). For each stimulation site, the strength of theconnection was calculated as the average current response (in pA)following stimulation. The connection strengths at each stimulationsite were used to generate a map of the monosynaptic connections toindividual POMC or NPY neurons, where each site corresponds to apixel position in the map (Fig. 1e–h). Anatomical boundaries weredetermined post-recording (Fig. 1i) and overlaid onto synaptic inputmaps. We distinguished POMC and NPY neuronal subtypes using micewith a bacterial artificial chromosome transgene containing either pomcor npy genomic sequence (plus B200 kb flanking sequence) modifiedto express green fluorescent protein (GFP) so that the cell type waslabeled and could be identified by fluorescence emission10 (Fig. 1j).Thus, by combining LSPS with cell type-specific GFP expression, wecould determine if neuronal microcircuits were specifically associatedwith molecularly defined neuronal cell types. Such distinct circuitsmight be predicted based on the opposing functions of POMC andNPY neurons. To represent the observed projections to POMC, NPYand unlabeled cells, we denote cell type (when known), anatomiclocation, and the sign of functional connectivity. For example,mVMHexc-ARCPOMC describes a connection in which the mVMHprojects excitatory synapses to POMC neurons in the ARC (see below).

Resolution of photostimulation mapping in the hypothalamus

LSPS is used to identify presynaptic neuron location by glutamatestimulation of action potentials in presynaptic neurons that results in apostsynaptic current response in the voltage-clamped neuron (Fig. 1f).The spatial distribution of uncaging sites that produce action potentialsin a presynaptic neuron defines the effective resolution for mapping theorigin of the synaptic responses. Because somatic and dendriticglutamate stimulation can generate action potentials, the resolutionwith which presynaptic neuron location can be determined depends, inpart, on both neuronal structure and membrane properties. Thus,control experiments are required to determine if there are differences inthe responsiveness to glutamate uncaging in the areas that will beinvestigated for synaptic mapping.

To determine the spatial distribution of uncaging sites around aneuron that elicited action potentials, we constructed excitation

d

b

ARC

VMH

LH

3V

0 100 200

UV

Time (ms)

i

ii

iii

200 pA

75 pA

3V

VMH

ARC

UV

Distance (µm)

3V

c

fe

hg

ji

a

3V

Distance (µm) –600 –400 –200 0 200 400 600

–600 –400 –200 0 200 400 600

pA

pA

0

10

20

30

40

–6

–4

–2

0

Figure 1 Laser scanning photostimulation for mapping hypothalamic neural

circuits. (a) Hypothalamic brain regions that were studied. 3V: third ventricle.

(b) Brain slice of mediobasal hypothalamus (300 mm thick). The boundaries

of the ARC and VMH are indicated. Red circles: laser stimulation sites. Green

circle: voltage-clamped soma. The patch pipette is marked with a dashed

black outline. Scale bar equals 300 mm. (c) Three types of responses

observed in LSPS. (i) Direct response. Dark gray bar marks 0–5 ms.

(ii) Synaptic response. Light gray shading marks 5–75 ms. (iii) No response.(d) Focal glutamate release near the soma or dendrite causes action potential

firing and neurotransmitter release (pink ball) which is detected as a

postsynaptic current (red). (e) Map of synaptic responses to photostimulation.

Colors represent mean synaptic current measured 5–75 ms after

photostimulation. Black pixels: direct responses. White square: soma

position. Dotted white lines define anatomical boundaries from individual

brain slices. (f) Examples of individual postsynaptic currents from black

boxed region of e. Scale bars: 35 pA, 25 ms. (g) From a different neuron,

inhibitory synaptic input map. (h) Examples of individual inhibitory

postsynaptic currents resulting from glutamate uncaging from black boxed

region of g. Scale bars: 35 pA, 125 ms. (i) Fluorescence micrograph (false-

color image) of a fixed brain slice stained with Sytox Green was used to

define nuclear boundaries, post-recording. Scale bar: 300 mm. (j) Voltage-

clamped neuron under infrared gradient contrast optics (top) and during

fluorescence emission (bottom, false-color image), identifying it as a POMC

cell. Scale bars: 15 mm.

NATURE NEUROSCIENCE VOLUME 8 [ NUMBER 10 [ OCTOBER 2005 1357

ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

profiles17 by recording evoked action potentials (loose-seal extracellularrecordings) after stimulation by focal uncaging of glutamate in a gridpattern of stimulation sites centered on the soma (Fig. 2a). The spatiallocation and the number of evoked action potentials were representedin a color map (Fig. 2b). Excitation profiles were generated for multipleneurons in three brain regions: ARC, medial VMH (mVMH) andlateral VMH (latVMH). Average excitation profiles (Fig. 2c,d) por-trayed the spatial distribution of glutamate responsiveness (the resolu-tion of photostimulation) in each of these areas. These excitationprofiles demonstrated that the strongest activation is at sites adjacent tothe soma, with a fall-off in responsiveness at more distant secondarysites. We quantified the resolution as the weighted mean distance (R)from the soma to each site of action potential stimulation (R ¼ S(r �n)/Sn, where r is the distance to the soma for each stimulation site andn is the number of action potentials at that site). The mean resolutionwas B140 mm for the ARC, mVMH and latVMH under fed and fastedconditions and for ob/ob mice (Fig. 2e). Thus, in synaptic maps, aconnected presynaptic neuron may be expected to generate a post-synaptic response not only at the location of its soma but also, onaverage, within two stimulation sites surrounding the soma, as seen inexcitation profiles. This resolution is less refined than that reportedin barrel cortex neurons (B50% greater than for layer 4 and layer5 neurons in ref. 17). This may be related to the very high membraneresistance in these hypothalamic neurons (typically 41 GO), whichmight be expected to enhance the excitability of these neurons fromstimulation at more distal dendritic sites if this results in a largermembrane length constant and increased electrotonic conduction.

Also, for synaptic maps, the magnitude of the response in thevoltage-clamped neuron is dependent, in part, on the number of actionpotentials generated in the presynaptic neuron in response to gluta-mate uncaging (glutamate excitability)17. The glutamate excitabilitywas B12–16 action potentials/cell for the ARC, mVMH and latVMH(Fig. 2f), and the differences were not statistically significant. Thus, insynaptic mapping, differences in the strength of inputs from theseanatomically separate regions (see below) were not attributable todifferences in either glutamate excitability or the resolution of gluta-mate uncaging.

We also established that LSPS, under the conditions used here,measures only monosynaptic connections between an uncaging siteand the voltage-clamped neuron (as opposed to activating a poly-synaptic chain of neurons). Excitation profiles with and withoutcalcium in the bath were used to confirm that synaptic input resultingfrom focal uncaging of glutamate was not sufficient to evoke actionpotentials. Comparison of excitation profiles recorded in B0 mMCa2+/8 mM Mg2+ with those for the same cell in 4 mM Ca2+/4 mMMg2+ did not show any additional sites of evoked action potentialsin neurons recorded from the ARC, mVMH and latVMH (datanot shown). Under low Ca2+, a condition under which synaptictransmission is improbable, the excitation profile could only representdirect stimulation of the recorded neuron with no contribution fromsynaptic input. We did not observe any additional sites of actionpotential generation when synaptic transmission was enabled bycalcium, indicating that synaptic input maps represent only mono-synaptic connections.

These control experiments demonstrated that LSPS is well-suited formapping distributions of monosynaptic inputs to hypothalamic neu-rons and that the mapping resolution was sufficient to permit pre-synaptic neuron location to be determined within subregions ofhypothalamic nuclei. Thus, using LSPS, we set out to identify func-tional connections from the VMH to ARC neurons involved inregulating energy homeostasis.

Excitatory synaptic input maps

We first surveyed the overall excitatory VMH inputs to the ARC byrecording from a set of unlabeled ARC neurons (n ¼ 16) in whichneuronal subtype was not identified. Unlabeled ARC neurons includedPOMC, NPY and other ARC neuronal subtypes. The stimulation siteswere recorded in the same orientation for different cells, allowing mapsof multiple cells to be averaged (example maps for individual cells inthis group are shown in Fig. 3, top). The resulting group input mapshowed the regions that gave rise to the strongest and most consistent

APs0

1

2

c

a b

d

e f

UV

600 µm

Mea

n di

stan

ce (

µm)

ARC mVMH latVMHCell location

200

150

100

50

0

AP

s/ce

ll

ARC mVMH latVMHCell location

20

15

10

5

0

ARC, control

n = 23

mVMH, control

n = 19

latVMH, control

n = 20

0

1

3

2

APs

0 300 600Distance (µm)

n = 13

ARC, fasted

n = 12

mVMH, fasted

n = 8

latVMH, fasted

Figure 2 Excitation profiles and resolution of stimulation by glutamate

uncaging. (a) Loose-seal extracellular recording of action potentials after

uncaging glutamate in a grid pattern around somata neuron defines the

spatial extent over which action potentials can be generated. (b) Action

potentials (APs) resulting from glutamate uncaging. The uncaging sites that

evoked action potentials are mapped and color-coded for the number of

action potentials. Scale bars equal 1 mV, 25 ms. (c,d) Average excitation

profiles for neurons in the ARC, mVMH and latVMH under (c) controlconditions and (d) in fasted mice. Scale bar: 150 mm. (e) Average resolution

of photostimulation. Resolution was determined by calculating the weighted

mean distance (R) to an action potential (R ¼ S(r � n)/Sn, where r is the

distance to the soma for each stimulation site and n is the number of action

potentials at that site) in excitation profiles from neurons in the ARC, mVMH

and latVMH. (f) Average excitability to glutamate uncaging. The glutamate

excitability of neurons in each region was calculated from excitation profiles

as the total number of action potentials measured for each cell (APs/cell).

Black bars: control conditions, gray bars: fasted conditions, white bars: ob/ob.

Error bars are s.e.m.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

excitatory presynaptic inputs to these ARC neurons. For unlabeledARC neurons, the excitatory synaptic input (Fig. 4a, left) was broadlydistributed across both the VMH and the ARC with the strongestconcentration of inputs coming from the latVMH (Fig. 4a, left,arrowhead). There were also excitatory inputs originating withinthe ARC as well as the mVMH, and inspection of input maps toindividual cells showed that a small subset of unlabeled neuronsreceived input predominantly from the mVMH (4/16). However,these were diluted in the average synaptic map for the entire sampleof unlabeled ARC neurons by the input distributions of the other cells.We refer to these VMH projections as VMHexc-ARCunknown, denotingthe lack of knowledge about the identity of the postsynaptic cell.These results are the first demonstration of direct, functional inputsfrom the VMH to the ARC, highlighting the usefulness of LSPS foridentifying new circuit elements. This sample of input distributionsonto unlabeled ARC neurons also provided a basis for comparisons tothe input distributions generated by recordings from molecularlydefined neurons.

We next focused on inputs to the POMC neurons in slices from pomc-gfp BAC transgenic mice (Fig. 4a, middle). The origin of excitatoryprojections to POMC neurons (n¼ 24) was concentrated as a ‘hotspot’,primarily within the mVMH (boxed in Fig. 4a, mean synaptic inputfrom boxed region: –4.2 ± 0.9 pA/stimulation site), with additional,sparse inputs from elsewhere in the mVMH, ARC and periventricularregions. A few neurons also received input from the latVMH (Fig. 3,middle). Projections from the mVMH represented a subset of theinput distribution evident when recording from unlabeled neuronsin the ARC, and the mVMHexc-ARCPOMC projection was signifi-cantly stronger than projections from the corresponding region in

mVMHexc-ARCunknown (Fig. 4c, Po 0.03, Mann-Whitney). This ob-servation suggests that in the studies of unlabeled neurons, the strengthof the mVMHexc-ARCPOMC projection was diluted by projectionsoriginating elsewhere and targeting other cell types that were repre-sented in the sample of unlabeled ARC neurons. Overall, these datashow that the mVMH contains a group of neurons that preferentiallyform excitatory synapses onto POMC neurons relative to other ARCneurons. This result also shows that recording from molecularly defined,labeled cell types can demonstrate specific spatial patterns of presynapticinput that are not evident when recording from unlabeled neurons.

The pattern of inputs to the arcuate NPY neurons was markedlydifferent from both the inputs to the POMC and unlabeled neurons.The average excitatory input to NPY neurons (n¼ 16) was significantlyless than for POMC or unlabeled ARC neurons (P o 0.02, Dunn’sTest). The spatial distribution was different than for POMC neurons,with the inputs originating from within the ARC and the latVMH(Fig. 4a, right, arrow). Many NPY neurons did not receive evokedsynaptic input, but a subset of the neurons (5/16) did show excitatoryinputs originating from the latVMH and adjacent lateral regions(Fig. 3, bottom). Notably, mVMH excitatory synaptic input to NPYneurons was absent, indicating that the mVMHexc-ARCPOMC andlatVMHexc-ARCNPY circuits are distinct. Note, however, that medialinputs to NPY neurons were observed in some slices from a morecaudal region containing the dorsomedial hypothalamic nucleusdirectly adjacent to the ARC (S.M.S. and J.M.F., unpublished data).Thus, although POMC and NPY neurons are interspersed in the ARC,they nevertheless receive anatomically distinct synaptic inputs.

Inhibitory synaptic input maps

We also mapped inhibitory synaptic inputs (Figs. 1i,j and 4b). POMCand unlabeled ARC neurons received the same total amount ofinhibitory synaptic input with similar spatial distributions (Fig. 4d,mean synaptic input from the ARC: POMC, 3.0 ± 0.7 pA/stimulationsite; unlabeled ARC, 3.6 ± 0.7 pA/stimulation site). Inhibitory inputsarose primarily from the ARC with only weak average inhibitory inputstrength from within the VMH. On the basis of ultrastructure analysis,POMC neurons receive inhibitory, GABA-containing synapses fromNPY neurons22, and the pattern of local ARC inhibition found here isconsistent with this. However, it is likely that other inhibitory cell typesare also involved. For example, we observed that the ventromedialARC, which has a high density of NPY neurons, was not a substantialsource of synaptic inhibition to the POMC and unlabeled neurons(Fig. 4b). The identity of the non-NPY cells projecting inhibitoryinputs onto the POMC neurons is not known.

We found only weak inhibitory synaptic projections to NPY neurons(Fig. 4b, right), and these originated almost exclusively from within theARC (Fig. 4d, mean synaptic input from the ARC: 1.0 ± 0.3 pA/stimulation site, P o 0.02, Dunn’s Test). This weak synaptic input toNPY neurons is consistent with a study that showed significantly lowersynapse density on NPY neurons relative to POMC neurons10. Notably,inhibitory synaptic input onto NPY neurons was not observed comingfrom the VMH. This observation mitigates the possibility that the VMHreduces food intake by directly projecting inhibitory synapses to NPYneurons and suggests that the VMH is more likely to contribute tosatiety by activating POMC neurons via mVMHexc-ARCPOMC ratherthan by inhibiting orexigenic NPY neurons.

Fasting markedly reduces mVMHexc-ARCPOMC

The presence of an excitatory circuit projection between the mVMHand POMC neurons (mVMHexc-ARCPOMC) suggests a possible rolefor this circuit in mediating the ability of the VMH to reduce food

0

–5

–10

–15pAU

nlab

eled

AR

CP

OM

CN

PY

0

–5

–10pA

0

–5

–10pA

Figure 3 Synaptic input maps for individual unlabeled ARC, POMC and

NPY neurons. Maps were repeated two to four times and averaged for each

neuron. Dotted white lines define anatomical boundaries from individualbrain slices. Black pixels: direct responses. White square: soma position.

Scale bar: 300 mm.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

intake. To correlate the strength of the mVMHexc-ARCPOMC projec-tion with changes in nutritional state, we mapped the synaptic inputs toPOMC neurons after food deprivation. If synaptic circuits have a rolein regulating feeding, the connection strength might be expectedto decrease in mVMHexc-ARCPOMC and/or increase in ARCinh-ARCPOMC. Mice were fasted for 24 h, over which time they lost 17 ±0.8% of their body weight. In fasted animals, there was a significantreduction in the strength of mVMHexc-ARCPOMC relative to thestrength of that projection in fed mice (mean synaptic input fromboxed region: –2.1 ± 0.6 pA/stimulation site, P¼ 0.03, Mann-Whitney;Fig. 5a). Control experiments ruled out differences in the intrinsicresponsiveness of mVMH neurons to glutamate stimulation afterfasting as a cause of this reduction (Fig. 2c–f). To visualize sites ofcircuit plasticity, we computed difference maps by subtracting theaverage excitatory synaptic input map for POMC neurons in fastedmice from that of fed mice. This showed a significant difference in thestrength of the mVMHexc-ARCPOMC circuit between these two states(Fig. 5d). In addition, a small but significant gain of spatially distinctexcitatory synaptic inputs to POMC neurons arising from the lateralARC/VMH boundary was observed (Fig. 5d,f, region of interest 4). Thetotal inhibitory synaptic inputs to POMC neurons from the ARC,ARCinh-ARCPOMC, were not significantly different in fasted micerelative to the fed state (Fig. 5g,i). These data show that the mVMHexcitatory connections to POMC neurons show marked plasticitywith food deprivation (Fig. 5f, regions of interest 1–3). The demon-stration that the mVMHexc-ARCPOMC projection is downregulatedby fasting is also consistent with the known role of the VMH as a satiety

center and suggests that this projection mayhave a functional role to increase feeding afterfood restriction.

Leptin does not block reduction of

mVMHexc-ARCPOMC

The response to fasting includes changes in thelevels of a number of hormonal and metabolicsignals including leptin, which decreases sig-nificantly with fasting23. To test whether thealtered inputs to the POMC neurons resultedfrom the decreased leptin concentration asso-ciated with food deprivation and weight loss,we treated fasted animals with leptin as pre-viously described23. In this case, average exci-tatory synaptic input strength of mVMHexc-ARCPOMC (Fig. 5c) was similar to that fromthe fasted animals that did not receive leptin

(mean synaptic input from mVMH hotspot: –1.3 ± 0.2 pA/stimulationsite, P 4 0.7, Mann-Whitney; also Fig. 5f, regions of interest 1–3).These data suggest that low leptin levels are unlikely to be directlyresponsible for the change associated with fasting in synaptic connec-tions between mVMH neurons and POMC neurons associated withfasting and that some other signal is responsible. Consistent with this,group excitatory synaptic input maps from leptin-deficient ob/ob,pomc-gfp mice (Fig. 5c,f) were not different from those from fed,wild-type pomc-gfp mice, further indicating that this projection is alsonot affected by leptin deficiency. Inhibitory inputs to POMC neuronswere also unchanged with leptin treatment of fasted mice (Fig. 5h,i).However, the site of increased excitatory input to POMC neurons fromthe latARC/VMH was no longer evident when fasted animals receivedleptin (Fig. 5d–f, region of interest 4). Thus, this small hotspot, which isidentified only after fasting, seems to be leptin regulated. The functionalrole of this projection is unclear, as it is not seen in other POMCsynaptic input maps, and on the basis of the sign of this connection, itwould be expected to paradoxically reduce food intake in fastedanimals. Alternatively, this projection could be related to other diversefunctions of POMC neurons, including secretion of b-endorphin,a neuropeptide that inhibits activity in gonadotropin-releasinghormone–expressing neurons in reproductive circuits24 and alsomodulates reward circuits25. Nevertheless, this finding highlights theunbiased nature of LSPS as a search tool for presynaptic neuron locationand that its use can identify projections in advance of an understandingof their function. Overall, these results establish that both fasting andleptin have dynamic effects on hypothalamic microcircuits.

*

**

UnlabeledARC

NPYPOMCUnlabeledARC

NPYPOMC

4

3

2

1

0

pA/s

timul

atio

n si

te

pA/s

timul

atio

n si

te

–6

–5

–4

–3

–2

–1

0ARC inhibitory hotspotmVMH excitatory hotspot

n = 17

n = 16 n = 24 n = 16

n = 21 n = 16

ARC

mVMH

latVMH

pA

pA

–4

–3

–2

–1

0

0

2

4

6

NPYPOMCUnlabeled ARCa

b

c d

Figure 4 Mean synaptic input maps for unlabeled

ARC, POMC and NPY neurons show the strengths

and distinct spatial distributions of synaptic inputs

onto these neuronal populations. (a) Excitatory

input maps. The black box defines the boundaries

of the mVMH hotspot. Arrowhead marks the

location of the concentration of inputs to

unlabeled ARC neurons; arrow denotes aconcentration of inputs to NPY neurons. White

squares mark soma location. Scale bar: 300 mm.

(b) Inhibitory input maps. (c,d) Mean synaptic

input to unlabeled ARC, POMC and NPY neurons

from (c) the mVMH excitatory hotspot and (d) from

the ARC inhibitory hotspot, which is defined by

the anatomical boundary of the ARC. *, P o 0.03,

Mann-Whitney comparison with unlabeled ARC.

**, P o 0.02, Dunn’s Test). Error bars are s.e.m.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

DISCUSSION

Physical2, chemical26 or genetic4 disruptions of the VMH cause obesity,and electrical stimulation of this nucleus causes a reduction of foodintake27. While these data establish an important role for this nucleus incontrolling food intake and body weight, the mechanisms by which theVMH regulates energy balance are largely unknown. In this report, weset out to test whether pathways exist through which the VMH mightregulate the activity of specific classes of ARC neurons that are knownto regulate food intake and body weight. We used LSPS to show afunctional interaction between the VMH and POMC neurons in theARC and identified an entirely new projection, mVMHexc-ARCPOMC.This projection carries excitatory inputs from the mVMH to POMCneurons in the ARC. Furthermore, we showed that this interaction israpidly and significantly modified by changes in nutritional state.Notably, the medial VMH subdivisions have been previously associatedwith regulation of energy balance and metabolism, with both leptin-28

and glucose-responsive neurons29 localized there. In aggregate, thesedata suggest a pathway by which the mVMH might regulate energybalance through excitatory synaptic input to POMC neurons thatsuppress feeding.

In principle, the VMH could also mediatesatiety by directly inhibiting NPY neurons.However, we did not find any evidence sup-porting this alternate pathway, as there wereonly sparse inputs to NPY neurons from theVMH, and these were spatially distinct excita-tory inputs from the latVMH, a subregion thatis known to regulate sexual behavior30,31. AsNPY has been shown to regulate fertility, onepossible role of latVMHexc-ARCNPY may beto control aspects of reproductive behaviorand neuroendocrine function. Although thesedata do not rule out the existence of mVMHconnections to NPY neurons, the frequency ofsuch connections is presumably far less thanfor POMC neurons.

Functional mapping showed markedlydifferent spatial distributions of VMH inputsonto NPY versus POMC neurons in the ARC.A comparison of the NPY and POMC inputmaps with those from unlabeled ARC neurons(VMHexc-ARCunknown) suggests that themVMHexc-ARCPOMC and latVMHexc-ARCNPY comprise only a subset of the totalexcitatory projections to the ARC, indicatingthat there are other VMH projections ontounidentified ARC cell types. Overall, thesedata add several new elements to the wiringdiagram of the VMH and the ARC (Supple-mentary Fig. 1) and demonstrate that differ-ent ARC neurons with diverse functions areintegrated into spatially distinct synaptic cir-cuits. These results highlight the added infor-mation that can be generated by using LSPS inanimals in which specific neuronal cell typescan be identified.

Functional considerations for synaptic

input maps

These results also illustrate the markedplasticity of hypothalamic circuits10,32,33 in

response to changes in nutritional state. The dynamic regulation ofmVMHexc-ARCPOMC by food restriction and weight loss is consistentwith a proposed anorexigenic function for this projection, and suggeststhat the hyperphagia that develops after food deprivation may result, inpart, from a reduction of excitatory inputs from the mVMH to POMCneurons. However, the relative contribution to energy homeostasis ofmVMHexc-ARCPOMC and the other microcircuits reported hereremains to be established. An important future challenge will be tocorrelate activity in these circuits with specific biological responsesin vivo. This will require the development of new approaches tomodulate neuronal activity in molecularly defined microcircuitsembedded within complex, poorly characterized brain regions. Theidentification of molecular markers for these mVMH neurons, such asneuropeptides, would allow the cell-specific targeting of novel genecassettes that alter neuronal function. Techniques that permit theactivation or silencing34,35 of specific projections are under develop-ment and could be thus be used to assess the contribution of thisspecific neural circuit to feeding behavior. Our results represent anecessary first step towards linking specific neural circuits and feedingbehavior by demonstrating distinct microcircuits to POMC and NPY

a c

pA/s

timul

atio

n si

te

Fed Fasted

b

d fe

g ih

Fed – fasted Fed – fast/leptin

POMC, fasted POMC, fast/leptin

POMC, fasted POMC, fast/leptin

12

34

5

n = 17 n = 16

POMC, ob/ob

n = 17

mVMH

latVMH

ARC

12

34

5n = 17

12

34

5

12

34

5n = 15

0

–1

–2

–3

–4

pA

6

4

2

0

pA

12

34

5Region of interest

pA

FedFastFast + leptin

#

** *

ob /ob

0

–2

–4

–62 3 41 5

0

1

2

3

4

ARC inhibitory hotspot

Fast +leptin

pA

0

–1.5

–3.0

1.5

3.0

Figure 5 Analysis of circuit plasticity after fasting and with leptin treatment during fasting. (a–c) Mean

excitatory synaptic input maps for POMC neurons in (a) mice after a 24 h fast, (b) fasted mice that

received leptin and (c) ob/ob mice. (d,e) Difference maps were made by subtracting (d) fed minus fasted

or (e) fed minus fasted with leptin treatment to illustrate sites of circuit plasticity. Negative values signify

decreased excitatory synaptic input onto POMC neurons relative to the level in fed animals, and positive

values signify increased excitatory inputs. Numbers along the diagonal gray dashed lines in a–e markregions of interest. (f) Average current strength from regions of interest in the maps from fed, fasted and

fasted + leptin-treated groups. #: P ¼ 0.07, *: P o 0.05, ANOVA. (g,h) Mean inhibitory synaptic input

maps for POMC neurons from mice after (g) 24 h fast and from (h) fasted mice that received leptin.

(i) Mean inhibitory synaptic input from the ARC hotspot. White squares mark soma locations. Scale bar:

300 mm. Error bars are s.e.m.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

neurons and correlating the strength of one such microcircuit(mVMHexc-ARCPOMC) with metabolic state.

Technical considerations for synaptic input maps

In these studies, we identified ‘hotspots’ of synaptic input that wereanatomically distinct for POMC and NPY neurons. These hotspotscorrespond to the location of neurons that project the strongest andmost frequently observed synaptic inputs for each cell type. Theinterpretation that the mVMHexc-ARCPOMC hotspot correspondsto presynaptic neuron location in the mVMH as opposed to dendriticstimulation of neurons in the ARC is based on three lines of evidence:(i) excitation profiles of ARC neurons show that maximal activation ofthe ARC neurons occurs near the soma (Fig. 2c), whereas in synapticinput maps onto POMC neurons, the strongest region of the hotspotextends well into the mVMH and not substantially into the ARC; (ii)inspection of synaptic input maps for individual POMC neurons showsthat sites of synaptic input are localized to the VMH and typicallydo not extend into the ARC and (iii) Golgi studies indicate that ARCneurons only rarely extend dendrites into the main body of the VMH13.Notably, the hotspots reported here for different neuron subtypesare spatially quite distinct, further demonstrating that thespatial resolution in these studies is sufficient to identify distincthypothalamic microcircuits.

An additional consideration is that the connections observed in thesestudies were restricted to intact projections within the brain slicepreparation. By recording input maps to unlabeled ARC neurons, wedetermined the extent to which intact circuit connections could bereliably obtained in this coronal slice plane (Fig. 4a,b, left). Weobserved robust connectivity from across the ventral surface of thisslice plane at both medial and lateral positions. There is clearly a fall-offin connectivity from sites at the dorsal VMH boundary that is likelydue to an increased probability of cut connections, and the resultingdata should not be interpreted as an absence of inputs from theseregions. For this reason, LSPS is best suited for analysis of specificmicrocircuits, as opposed to a global analysis of all of the afferent inputsto a brain region. The focus of this study was to test functionalconnectivity between the VMH and the ARC and to dissect the originsof inputs from the VMH onto molecularly defined neurons in the ARC.For this aim, functional mapping is strong evidence for connectivity,which, in the case of VMH-ARC, had not been fully tested based onanatomical methods. It can be expected that glutamate uncaging inother brain slice orientations will demonstrate additional patterns ofsynaptic input to the molecularly defined neurons in the ARC.

In summary, we have used LSPS to map functional connectionswithin the hypothalamus and identified a new excitatory circuitbetween the mVMH and POMC neurons in the ARC (mVMHexc-ARCPOMC). By applying LSPS to molecularly defined, GFP-labeledneurons, we have observed a more refined spatial distribution ofprojections than was evident when using unlabeled neurons. We havealso demonstrated that POMC and NPY neurons, which are inter-spersed in the ARC but have opposing effects on feeding, receivepresynaptic inputs from anatomically distinct regions. The observedexcitatory projection of mVMH neurons to anorexigenic POMCneurons and its diminution by fasting provide the basis for thehypothesis that the VMH reduces food intake by activating POMCneurons, a possibility that remains to be tested in vivo. Conversely, theabsence of inhibitory inputs to NPY neurons from the VMH suggeststhat it is less likely that the VMH might promote satiety by directlyinhibiting the activity of NPY neurons. Finally, we show that the use ofLSPS in animals with GFP-labeled neurons allows the generation ofdetailed wiring diagrams of feeding circuits (Supplementary Fig. 1).

The approach used here may also prove valuable for identifying newmolecularly defined neural circuits controlling other behaviors.

METHODSSlice preparation, electrophysiology, glutamate uncaging. Experimental tech-

niques were similar to those reported previously17, and only the differences are

described here. Animal care and experimentation followed Rockefeller Uni-

versity’s Care and Use guidelines. Only male mice (P23–P33) were used in this

study. Fasted mice were deprived of food for 24 h. For one group of fasted

animals, leptin (1 mg/g), was administered intraperitoneally (i.p.) at the start of

the fast and after 12 h. Brain slices were always prepared within 1 h before the

end of the dark cycle. Typically, one coronal slice (300 mm thickness) per animal

was cut using a vibratome in chilled cutting solution containing (in mM): 110

choline chloride, 25 NaHCO3, 11 glucose, 2.5 KCl, 1.25 NaH2PO4, 11.6 sodium

ascorbate, 3.1 sodium pyruvate, 7 MgCl2, 0.5 CaCl2, aerated in 95% O2/5%

CO2. Slices were transferred to artificial cerebrospinal fluid (ACSF) containing

(in mM): 119 NaCl, 25 NaHCO3, 11 glucose, 2.5 KCl, 1.25 NaH2PO4, 4 MgCl2,

4 CaCl2, aerated in 95% O2/5% CO2. High divalents were used to reduce

spontaneous synaptic activity. The slice was incubated at 34 1C for B30 min

and then transferred to a recording chamber containing ACSF with 0.24 mM

caged, MNI-glutamate (4-methoxy-7-nitroindolinyl L-glutamate, Tocris) at

22 1C. Neurons were identified by fluorescence emission and then visually

targeted with infrared gradient contrast optics. Neurons 435 mm from the

surface of the slice were patched using electrodes with tip resistances 3.5–5 MO.

The intracellular solution contained (in mM): 132 cesium trifluoromethane-

sulfonate, 7 CsCl, 10 HEPES, 4 Mg2ATP, 0.3 Na3GTP, 10 sodium phospho-

creatine, 3 sodium ascorbate, 1 EGTA, 0.05 Alexa-594 (Molecular Probes).

Series resistances were 13–30 MO. Input resistances were 41 GO. Excitatory

postsynaptic currents were isolated at the holding potential –62 to –58 mV.

Inhibitory postsynaptic currents were isolated at the reversal potential of the

glutamate-activated current (0 to +12 mV). Laser stimulus was 1 ms.

Data analysis. Data analysis was similar to that reported previously17, and only

the differences are described here. Responses were analyzed for 75 ms after the

ultraviolet stimulus for excitatory PSCs and 250 ms for inhibitory PSCs owing

to their larger amplitude and slower decay time. The first 5 ms was parsed into

direct responses, and the remaining time was the synaptic window. Typically,

two to four maps were obtained for each cell and were averaged, reducing noise

due to spontaneous PSCs. Note that the contribution of spontaneous PSCs was

small, and these magnitudes were not statistically different between sample

populations. These averaged single maps were then averaged across all of the

cells recorded into a group map for each cell type or experimental condition.

This was done by aligning the uncaging pattern to the base of the brain and the

third ventricle so that maps were recorded with the same orientation. Post-

recording, slices were fixed, stained with 1% Sytox Green (Molecular Probes) in

phosphate buffer/0.3% Triton-100, and fluorescence emission was used to

visualize anatomical boundaries (Fig. 1i). Images were overlaid with brightfield

images of the slice in the recording chamber using the third ventricle, base of

the brain and other anatomical markings for alignment. In this way, anatomical

boundaries of each slice could be mapped onto specific pixels in the synaptic

input maps. In group maps, average anatomical boundaries were calculated for

display purposes. Also, for display only, the averaged group maps were filtered

with a two dimensional Gaussian (sigma: 49 mm, tiling: 3 � 3) to smooth the

maps. Data collection and data analysis were performed using software written

in Matlab (Mathworks).

Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTSWe are grateful to K. Svoboda for his generous assistance in implementing LSPS(funding to K.S. from the Howard Hughes Medical Institute (HHMI)) and toX.-L. Cai for assistance with mouse breeding. Support by Helen Hay WhitneyFoundation (S.M.S.), HHMI (G.M.G.S., J.M.F.) and the US National Institutesof Health (J.M.F.).

COMPETING INTERESTS STATEMENTThe authors declare that they have no competing financial interests.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Published online at http://www.nature.com/natureneuroscience/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Kennedy, G.C. The hypothalamic control of food intake in rats. Proc. R. Soc. Lond. B137, 535–549 (1950).

2. Hetherington, A.W. & Ranson, S.W. Hypothalamic lesions and adiposity in the rat. Anat.Rec. 78, 149–161 (1940).

3. Anand, B.K. & Brobeck, J.R. Hypothalamic control of food intake in rats and cats. Yale J.Biol. Med. 24, 123–140 (1951).

4. Majdic, G. et al. Knockout mice lacking steroidogenic factor 1 are a novel genetic modelof hypothalamic obesity. Endocrinology 143, 607–614 (2002).

5. Elmquist, J.K., Elias, C.F. & Saper, C.B. From lesions to leptin: hypothalamic control offood intake and body weight. Neuron 22, 221–232 (1999).

6. Clark, J.T., Kalra, P.S., Crowley, W.R. & Kalra, S.P. Neuropeptide Y and humanpancreatic polypeptide stimulate feeding behavior in rats. Endocrinology 115,427–429 (1984).

7. Zarjevski, N., Cusin, I., Vettor, R., Rohner-Jeanrenaud, F. & Jeanrenaud, B. Chronicintracerebroventricular neuropeptide-Y administration to normal rats mimics hormonaland metabolic changes of obesity. Endocrinology 133, 1753–1758 (1993).

8. Huszar, D. et al. Targeted disruption of the melanocortin-4 receptor results in obesity inmice. Cell 88, 131–141 (1997).

9. Fan, W., Boston, B.A., Kesterson, R.A., Hruby, V.J. & Cone, R.D. Role of melanocorti-nergic neurons in feeding and the agouti obesity syndrome. Nature 385, 165–168(1997).

10. Pinto, S. et al.Rapid rewiring of arcuate nucleus feeding circuits by leptin. Science304,110–115 (2004).

11. DeFalco, J. et al. Virus-assisted mapping of neural inputs to a feeding center in thehypothalamus. Science 291, 2608–2613 (2001).

12. Canteras, N.S., Simerly, R.B. & Swanson, L.W. Organization of projections from theventromedial nucleus of the hypothalamus: a Phaseolus vulgaris-leucoagglutinin studyin the rat. J. Comp. Neurol. 348, 41–79 (1994).

13. van den Pol, A.N. & Cassidy, J.R. The hypothalamic arcuate nucleus of rat–a quantitativeGolgi analysis. J. Comp. Neurol. 204, 65–98 (1982).

14. Callaway, E.M. & Katz, L.C. Photostimulation using caged glutamate reveals functionalcircuitry in living brain slices. Proc. Natl. Acad. Sci. USA 90, 7661–7665 (1993).

15. Sawatari, A. & Callaway, E.M. Convergence of magno- and parvocellular pathways inlayer 4B of macaque primary visual cortex. Nature 380, 442–446 (1996).

16. Dantzker, J.L. & Callaway, E.M. Laminar sources of synaptic input to cortical inhibitoryinterneurons and pyramidal neurons. Nat. Neurosci. 3, 701–707 (2000).

17. Shepherd, G.M., Pologruto, T.A. & Svoboda, K. Circuit analysis of experience-dependentplasticity in the developing rat barrel cortex. Neuron 38, 277–289 (2003).

18. Brivanlou, I.H., Dantzker, J.L., Stevens, C.F. & Callaway, E.M. Topographic specificity offunctional connections from hippocampal CA3 to CA1. Proc. Natl. Acad. Sci. USA 101,2560–2565 (2004).

19. Yoshimura, Y., Dantzker, J.L. & Callaway, E.M. Excitatory cortical neurons form fine-scale functional networks. Nature 433, 868–873 (2005).

20. Shepherd, G.M., Stepanyants, A., Bureau, I., Chklovskii, D. & Svoboda, K. Geometricand functional organization of cortical circuits. Nat. Neurosci. 8, 782–790 (2005).

21. Callaway, E.M. Cell type specificity of local cortical connections. J. Neurocytol. 31,231–237 (2002).

22. Cowley, M.A. et al. Leptin activates anorexigenic POMC neurons through a neuralnetwork in the arcuate nucleus. Nature 411, 480–484 (2001).

23. Ahima, R.S. et al. Role of leptin in the neuroendocrine response to fasting. Nature 382,250–252 (1996).

24. Faletti, A.G. et al. beta-Endorphin blocks luteinizing hormone-releasing hormone releaseby inhibiting the nitricoxidergic pathway controlling its release. Proc. Natl. Acad. Sci.USA 96, 1722–1726 (1999).

25. Kreek, M.J. Methadone-related opioid agonist pharmacotherapy for heroin addiction.History, recent molecular and neurochemical research and future in mainstreammedicine. Ann. NY Acad. Sci. 909, 186–216 (2000).

26. Shimizu, N., Oomura, Y., Plata-Salaman, C.R. & Morimoto, M. Hyperphagia and obesityin rats with bilateral ibotenic acid-induced lesions of the ventromedial hypothalamicnucleus. Brain Res. 416, 153–156 (1987).

27. Stenger, J., Fournier, T. & Bielajew, C. The effects of chronic ventromedial hypothalamicstimulation on weight gain in rats. Physiol. Behav. 50, 1209–1213 (1991).

28. Elmquist, J.K., Bjorbaek, C., Ahima, R.S., Flier, J.S. & Saper, C.B. Distributions ofleptin receptor mRNA isoforms in the rat brain. J. Comp. Neurol. 395, 535–547(1998).

29. Ono, T. et al. Glucoresponsive neurons in rat ventromedial hypothalamic tissue slices invitro. Brain Res. 232, 494–499 (1982).

30. Christensen, L.W., Nance, D.M. & Gorski, R.A. Effects of hypothalamic and preopticlesions on reproductive behavior in male rats. Brain Res. Bull. 2, 137–141 (1977).

31. Pfaff, D.W. & Sakuma, Y. Deficit in the lordosis reflex of female rats caused by lesions inthe ventromedial nucleus of the hypothalamus. J. Physiol. (Lond.) 288, 203–210(1979).

32. Theodosis, D.T. & Poulain, D.A. Activity-dependent neuronal-glial and synaptic plasti-city in the adult mammalian hypothalamus. Neuroscience 57, 501–535 (1993).

33. Bouret, S.G., Draper, S.J. & Simerly, R.B. Trophic action of leptin on hypothalamicneurons that regulate feeding. Science 304, 108–110 (2004).

34. Slimko, E.M., McKinney, S., Anderson, D.J., Davidson, N. & Lester, H.A. Selectiveelectrical silencing of mammalian neurons in vitro by the use of invertebrate ligand-gated chloride channels. J. Neurosci. 22, 7373–7379 (2002).

35. Lechner, H.A., Lein, E.S. & Callaway, E.M. A genetic method for selective and quicklyreversible silencing of Mammalian neurons. J. Neurosci. 22, 5287–5290 (2002).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

topographic mapping of vmh → arcuate nucleus microcircuits and their reorganization by fasting

Documents